Re: Happy New Year, CAP!

On Sun, 01 Jan 2006 01:02:33 GMT, Risujin <risujin@xxxxxxxxxxx> in
comp.ai.philosophy wrote:

>Traveler wrote:
>>>Just out of curiosity but, show of hands, how many here are actually
>>>working on "true AI"? (And I mean more than philosophical implications,
>>>but real theory.)
>>
>> Well I, for one, have been working on true AI from the very beginning
>> (early 1990s) although I'm now considered a crank and a crackpot by
>> many and I don't blame them. Still, I do have a few supporters.
>> ahahaha...
>>
>> BTW, true AI, to me, means any AI mechanism that can be scaled up to
>> human-level intelligence by adding more processing power.
>
>I don't buy the "more processing power" paradigm. If you can describe
>exactly how the scaled-up version will emerge into the desired behavior
>then you could simply describe the scaled-up version in the first place.

This is a reasonable objection. But it also neglects a couple of
important considerations. Scaling up intelligence mechanisms
obviously requires more processing power. The real difficulty with
any elementary intelligence mechanism is that it wouldn't look very
intelligent and conventional empirical standards for the production
of ai require that an ai artifact look and act intelligent. So there's
no easy way to judge the presence of intelligence in elementary
mechanisms without the appearance of intelligent behavior requiring
substantial computing power. We could only judge the presence of
intelligence on theoretical grounds but we have no way to do that
without some theory of intelligence to begin with and that is exactly
what is lacking.

This is exactly what is so stupid about academic rejection of theories
and the analysis of intelligence in favor of robotics and declarations
of intelligence. If we look at human embryos and even infants there is
no obvious indication of consciousness even though we can reasonably
expect its manifestation over time as the organism develops. Academic
empirics simply demand organics manifest overt conscious behavior
immediately.

So the "processing power" requirement does have some justification.
The real difficulty entails the means to judge whether what we have at
the level of elementary mechanisms is intelligence to begin with which
would justify added computing power. The claims of academic robotics
don't address this issue since they only recognize superficial aspects
of intelligent behavior. So I definitely agree scaling up appearances
alone in the absence of understanding intelligence is pointless.

The second objection to your comment is that it assumes intelligent
behavior at the scaled up level of "desired behavior" is already well
enough defined to act as a criterion for intelligent behavior. I think
if we understood what the scaled up macro behavior really is we
wouldn't have the problem with the micro intelligence mechanisms to
scale up. The most obvious problem with scaling arguments is they
aren't supposed to produce effects of intelligence until the scaling
is completed and we have no ability to project when the scaling will
be completed since every projection so far has been met with further
scaling demands.

>Whenever AI researchers say "oh it just needs more CPUs!" they simply
>acknowledge that their work is inadequate and that only if a million
>monkeys worked with a million of their thing for a million years would
>it work. This ridiculous argument even works for random command generation!
>
>> Temporal Intelligence:
>> http://www.rebelscience.org/AI/Temporal_Intelligence.htm
>
>I read your article and I don't quite understand what you mean by temporal.
>
>You dismiss the idea that the brain functions as a symbolic system, but
>later refer to signal discretization (into "spikes") within the brain.
>If signals are discretized, isn't this hinting at a symbol system?
>
>You assume that intelligent behavior can be broken down into "pieces of
>intelligence" that together produce emergent behavior/function (the
>connectionist approach). This in itself is not a given. Taking out a
>piece of this kind of system would reduce its capacity, whereas taking a
>piece out of, say, a computer would break it. How do we know which is
>correct?
>
>>>I've scoured the internet for papers, web-sites, anything and found
>>>nothing but hacks and phonies. (See: Singularity Instutute for one.)
>>
>> There are a few sites/people who claim to be working on true AI. Off
>> the top of my head, I can think of Jeff Hawkins'
>> www.onintelligence.org, Peter Voss' http://adaptiveai.com/ and a few
>> more that I can't remember right now.
>
>I read Hawkins' book, all he says is that we should study the brain in
>order to understand the intelligence mechanism as it appears in humans
>and work from there to reproduce it in an algorithm. He sprinkles in a
>few observations about certain neuron functions and draws
>generalizations which ultimately lead nowhere.
>
>As appealing as his vision of reverse engineering the brain sounds, its
>the wrong way to go. Medical science has been trying to do the same to
>much simpler human organs for centuries. Modern drugs are discovered
>largely by statistical studies. When you're forced to take some
>substance and test it to see if it has any useful properties by actually
>applying it to living tissue repeatedly to see what happens, you simply
>don't know *** about *how* it works.
>
>Peter Voss' venture strikes me as a money sink. He's just luring
>investors for zero return. If they really had made "significant
>progress", where are the publications and press releases?
>
>-- Risujin


~v~~

.

Loading